datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
multimodal-example
Multimodal Example Dataset
Small example dataset for testing multimodal (vision-language) fine-tuning with ms-swift.
Structure
├── train.jsonl # 10 training samples
├── test.jsonl # 2 validation samples
├── images/ # All referenced images (400x300 JPEG)
│ ├── dog_portrait.jpg
│ ├── forest_river.jpg
│ ├── laptop_desk.jpg
│ ├── mountain_lake.jpg
│ ├── ocean_rocks.jpg
│ ├── coffee_cup.jpg
│ ├── bookshelf.jpg
│ ├──… See the full description on the dataset page: https://huggingface.co/datasets/f13rnd/multimodal-example.ExamGeo
ExamGeo
ExamGeo contains 1,000 English plane-geometry problems for multimodal evaluation. Each record includes a problem statement, one diagram, a reference answer, and a difficulty value.
Contents
data.jsonl 1,000 problems
images/ 1,000 diagrams
Each line in data.jsonl has this format:
{
"id": "3",
"difficulty": "0.2",
"question_list": [
{"type": "text", "text": "..."},
{"type": "image_path", "image_path": "images/3_q0.png"}
],
"answer":… See the full description on the dataset page: https://huggingface.co/datasets/pH202411/ExamGeo.Boat_unity_exampleprompt2model-examples
Prompt2Model Toy Examples
Product: Prompt2Model:
a language-guided vision model factory. A typed pipeline (prompt, dataset config, training,
calibration/conformal abstain, ONNX export, an optional distill/quantize step with an
accuracy-floor gate, and a hard-case flywheel).
What this is (and isn't)
This is not a benchmark dataset. Prompt2Model has no natural "own" benchmark corpus the way a
task-specific product does. What's uploaded here is the repository's own… See the full description on the dataset page: https://huggingface.co/datasets/Dhi-Technologies/prompt2model-examples.Boat_unity_exampleBoat_unity_examplebengali-exams-public